In this paper we represent a multi-scale method for the detection of small
targets embedded in noisy background. The multi-scale representation is bui
lt using a weighted undecimated discrete wavelet transform. The method, in
essence, is based on the maximisation of information available at each reso
lution level of the representation. We show that such objective can be achi
eved by maximising Renyi's information. This approach allows us to determin
e an adaptive threshold useful for discriminating, at each scale, between w
avelet coefficients representing targets and those representing background
noise. Eventually, avoiding inverse transformation, scale-dependent estimat
es are combined according to a majority vote strategy. The proposed techniq
ue is experimented on a standard data set of mammographic images.